Rewrite workflows:review with context-managed map-reduce architecture #157
+462
−381
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Replace the v1 review command with a context-managed v2 that prevents context overflow by having sub-agents write findings to .review/ on disk and return only single-sentence summaries to the parent. Key changes: - Add PR Intent Analysis phase (shared context for all specialists) - Agents write JSON to .review/, return ~100 tokens each to parent - Add validation step to catch silent agent failures and invalid JSON - Add Judge phase for dedup, hallucination removal, and ranking - Add Deep Analysis phase for P1/P2 enrichment - Parent reads only ENRICHED_FINDINGS.json (~8k tokens vs ~30-50k in v1) - Smart agent selection based on PR content instead of running all agents - Cross-platform safe: uses project-relative .review/ instead of /tmp/ 🤖 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
…exploration - Add git-history-analyzer and code-philosopher to conditional agent selection - Expand Deep Analysis Phase with detailed stakeholder impact analysis (Developer, Operations, End User, Security, Business perspectives) - Add comprehensive scenario exploration checklist (invalid inputs, concurrency, scale, network, resource exhaustion, security vectors, etc.) - Note that deep analysis inherits the Ultra-Thinking approach from v1 but scoped to judged findings in an isolated context window 🤖 Generated with Claude Code Co-Authored-By: Claude <noreply@anthropic.com>
|
I ran into this issue today. I asked for a PR review, and it hit the limit. I am on the Max 5 plan. |
XSAM
reviewed
Feb 12, 2026
|
|
||
| Ensure that the code is ready for analysis (either in worktree or on current branch). ONLY then proceed to the next step. | ||
|
|
||
| - [ ] If ALREADY on the target branch → proceed with analysis on current branch |
Contributor
There was a problem hiding this comment.
I wonder why we keep - [ ] instead of a numbered list. What is the benefit of using - [ ] in this case?
7 tasks
- Remove 4 ghost agents (rails-turbo-expert, dependency-detective, devops-harmony-analyst, code-philosopher) that don't exist in plugin - Add 5 missing real agents to selection matrix (kieran-typescript-reviewer, kieran-python-reviewer, julik-frontend-races-reviewer, schema-drift-detector, learnings-researcher) - Replace Python3 JSON validation with Node.js for cross-platform compat - Add excluded paths check to judge prompt (not just agent prompts) - Add 0-findings short-circuit for clean PRs - Fix .gitignore append duplication with grep -q guard - Update frontmatter description to match v2 purpose - Bump version to 2.31.0
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
.review/on disk and return only ~100 token summaries to the parent, keeping parent context under ~12k tokens (vs ~30-50k in v1).review/instead of/tmp/(fixes Windows path issues)Key Architecture Changes
.review/, return 1 sentence each (~100 tokens vs 2-4k)ENRICHED_FINDINGS.json(~8k tokens max), spawns todo-creation agentsTest plan
/workflows:reviewon a PR with mixed file types (Ruby + JS) to verify agent selection.review/directory is created with expected agent output files.review/path works